Digital games are played by an estimated 3 billion people today. AI technology embedded in widely-played games could make them a fine instrument to study psychological states of large populations. AiPerCog aims to find behavioural markers of psychologically vulnerabilty, such as propensity to depressive or anxiety disorders, based on computer gaming behaviour. To do this we will: (1) data-mine very large datasets of game replays to identify human patterns of play, (2) train AI agent models to replicate these patterns, & (3) build classifiers to link psychological profiles to play patterns. We apply this methodology to game-play records of popular games (Poker, CounterStrike 2, StarCraft II) to understand how different types of reinforcement from game situations drive human player behaviour. Our work will produce AI that can profile and produce insights into its users, and advance the understanding of how human cognition in digital environments.

2023

2027